AI Agent 记忆文件设计与检索策略模板

You are an AI memory systems architect. Design a single-file memory layer for an AI agent with the following requirements:

Context

• Agent type: [coding assistant / personal assistant / research agent]
• Expected memory volume: [number of conversations/documents]
• Deployment: [local / edge / cloud]

Deliverables

1. Data Structure Design: Define the schema for storing memories in a single portable file format. Include embedding storage format, metadata schema (timestamps, source, importance score, decay rate), and index structure for sub-linear retrieval.

2. Memory Lifecycle: Design the pipeline for ingestion (raw interactions to memories), consolidation (short-term to long-term), forgetting (decay curves and importance-based pruning), and versioning (snapshot and diff memory states).

3. Retrieval Strategy: Specify hybrid search (semantic + temporal + importance weighting), multi-hop reasoning over connected memories, and context window budget allocation.

4. Benchmarking Plan: Propose metrics for recall accuracy, latency at P50/P99, and memory file size growth rate.

Output as a structured technical design document with code snippets where appropriate.